Dynamic data modeling based on Kalman state estimation

In the process of conventional historical data-driven modeling, when the initial state of the system at the starting point of the selected modeling data is unsteady, the model of the system cannot be accurately obtained. In addition, for many control systems, due to the mismatch between the control system performance and the controlled object or the frequent change of the control system setpoint, the system is in an unstable state for a long time, and it is difficult to manually screen the steady-state data. To solve this problem, a dynamic historical data modeling method based on Kalman filter is proposed. The dynamic data is divided into three sections. The first section is used for Kalman state estimation, the second section is used for model identification, and the third section is used for model verification. The results show that the model identification accuracy of this modeling method is high by using dynamic data, and the effectiveness of this method is proved.